search

peace195 / tensorflow-lite-YOLOv3

YOLOv3: convert .weights to .tflite format for tensorflow lite. Convert .weights to .pb format for tensorflow serving
106 stars 24 forks source link

Performance may be reduced when using .pb and .tflite tensorflow model #1

Open peace195 opened 4 years ago

peace195 commented 4 years ago

I need helps to review the YOLOv3 tensorflow version: https://github.com/peace195/tensorflow-lite-yolo-v3/blob/master/yolo_v3.py

It is the cause of reducing performance when using .pb and .tflite compare with darknet .weights.

peace195 commented 4 years ago

Please test the performance and give me your feedback.

nscotto commented 4 years ago

Thanks! Will do feedback as soon as I can.

titanbender commented 4 years ago

I've tested the models before and after conversion and I'm seeing significant worse results with the tf lite model. I'm wondering if the parameters are not configured correct.

peace195 commented 4 years ago

@titanbender Thanks. Let me investigate and fix it 👍

balajib363 commented 4 years ago

@titanbender I converted yolov3 tiny model to tflite but I am doing something wrong when i build in android app using tensorflow. Could guide with steps tobe followed before building in mobile app?? Thanks

franck-armand commented 4 years ago

After getting the .pb file, do you necessarily need to set the Docker env to get the .tflite ? also how should I set it? I am not quite familiar with how it works. thanks